Redefining AI: From Automation to Adaptive Systems with Daniel Hulme, Chief AI Officer at WPP

0:00The definition of stupidity is doing the same thing over again and expecting a different answer. So I would argue that by definition, automation is stupid. Now that it's not valuable, automation can be incredibly valuable. But for me, the true paradigm of AI are systems that can safely adapt themselves in production.

0:18Welcome to Data and AI Mastery, the podcast where we bring you cutting edge insights, practical advice and inspiring stories from the leaders shaping the future of data and AI across the globe. I'm your host, Raoul Gabriel-Urmer, founder of Cambridge Spark, the leader in transformational data and AI upskilling, career development and progression. In each episode, I will be diving into real-world case studies of companies harnessing the power of AI to drive innovation, reduce costs and create new business opportunities. So whether you are an aspiring data scientist, AI engineer or seasoned executive, this show is designed to give you the tools and knowledge to stay ahead in a world where data is transforming every aspect of business.

1:02Stay ahead, stay inspired, stay masterful. Welcome to Data and AI Mastery. Hi Daniel, great to see you. How are you doing today? Likewise, yeah, really, really good. Busy, busy week so far, but all good busy. Yeah, no doubt. In the world of AI, every week must be really busy as it stands. It is, yeah. It's relentless at the moment, but again, it's all going in the right direction.

1:30So if I take you to the role of Chief AI Officer, it's a fairly new-ish role, right? Like we kind of hear coming up. Can you tell us a bit, what does it involve essentially on the day-to-day and for WPP? Like, what is it all about? marketing is being completely disrupted by AI because you know in a world where you can now create ads in in seconds and test those ads against synthetic audiences as opposed to months then it completely changes our business model so my day job is is understanding how to apply AI to solving and differentiating the BPP's marketing platform which we call WP Open so I lean into problems like how we use AI to do dynamic segmentation or audience perception, content creation, channel optimization.

2:18So lots of different flavors of algorithms that I really enjoy thinking about and applying. Actually, WP also have a workforce of over a hundred thousand people. So how AI is going to be used to enhance, accelerate creativity and change our workforce. We still get to also build AI solutions and innovations for things outside marketing as well, which is also very exciting and keeps us close to other industries. And as part of that role, I also am a task with thinking about the kind of responsible, safe and ethical use of these technologies, not even just in this industry, but in other industries as well.

2:57So it's been a fantastic acquisition. They give me a huge amount of rope and resources to experiment with cool things.

3:10We talked about AI and I was watching your TEDx talk from four years ago, which is absolutely amazing. So I recommend everyone to watch it. And in that talk, you are describing, you know, what is AI? And I think there's two definitions. And one definition, you say, well, you know, there's a narrow definition. People talk about AI being like human-like capabilities, but maybe a bit too narrow. The second definition is where I'm more goal-oriented, you set a goal, AI can get there and learn by doing and learn through mistake. Can you elaborate a little bit more, you know, four years later in the world we're living today, you know, has that definition evolved, your sentiment evolved and how do you see it, you know, I guess in the world of business?

3:57Yeah, I think, you know, unfortunately that, you know, there are many definitions of eye knocking around. The most popular definition is, is I think the weakest definition and actually that's been popularized because of generative AI. So the most popular definition is getting computers to do things that humans can do. And now we've got, you know, AI's or generative AI creating corresponding natural language and recognizing objects and images. And when we make a machine to behave like humans, because humans are the most intelligent thing we know in the universe, we attribute that to be intelligent.

4:27Now there's enough books and information out there to show you that actually humans are not intelligent in certain ways. You know, we, we can find patterns about four dimensions. We can solve problems up to seven. Um, you know, computers can, can find patterns in thousands of dimensions and they can solve problems with thousands of moving parts. So, um, there's actually a much better definition of AI that I subscribe to this from the eighties, which is this goal directed adaptive behavior. So it's, then see systems are able to make decisions, learn about whether those decisions are good or bad, adapt themselves.

4:55So next time they make better decisions. And if I'm honest, most systems still in production are automation. They make decisions, but they don't learn from them. The definition of stupidity is doing the same thing over again and expects a different answer. So I would argue that by definition, automation is stupid. Now that it's not valuable, automation can be incredibly valuable. But for me, the true paradigm of AI are systems that can safely adapt themselves in production. And that's very hard to build, by the way. And so actually, whilst I aspire to building systems that do that, that fit that goal-directed adaptive definition, over the past several years, I've been looking at AI, not through definitions, not through technologies, but I've been developing frameworks to help organizations understand what frictions they have, and then how do they apply the right technologies to solving those frictions.

5:45So come at it much more from an application perspective and from a definition perspective. That's fantastic. So, you know, we've got this definition of intelligence and we can argue about it. And there's the feel of reinforcement learning, there's a feel of machine learning, but it sounds like I didn't there, you know, in the business world, let's start with the problem. Let's start with the friction point. And then there might be like simple solution that leverage automation without all the adaptability. So if we know it sounds obvious, but the reality is, is that industry, we get very excited about emerging technologies.

6:15Eight years ago, nine years ago, it was data science. So people would hire machine learning experts that hope that by extracting insights from data that it would lead to better decisions. And for the most part, it doesn't. Now it's obviously generative AI, but we then try and apply those emerging technologies to solving the wrong problems. We blame the technology. We say there's hype, but the reality is that humans really are not very good at understanding what the right algorithms are solving the right problem. And I think that's what Satalia and that's what we're really good at. Why do you think the industry gets it wrong in that case?

6:48Why do we get so excited about emerging technology and we forget about solving problems? Well, if I was being controversial, which I will be, people like to build careers around emerging technologies and they'll convince themselves and convince their stakeholders that by giving them some funding and building a data science team or building a generative AI team, that they're going to be able to then drive value. And because they want to kind of build a career around these exciting new emerging technologies. And the reality is that for the most part, most organizations can't attract and retain the talent to build differentiated solutions.

7:26That's really one blocker. The second blocker is if you've never built and scaled software inside your organization, then you're not going to do the same thing for AI. It's going to be very hard for you to build and scale and innovate that software. So I try to encourage people to focus on applying these technologies to solving problems in their supply chain that will differentiate their business. Core productivity will be solved in the future. Don't worry about building agents that do expenses and things like that. Just focus, focus on applying these technologies that differentiate your business.

7:58But the fact is people tend to get excited about this and build careers around these emerging technologies. hmm i love a bit of controversy then so that's great to uh to go there and um it sounds like fundamentally we have to go back to strategy your value proposition what's different about your business and kind of double down on what's the differentiator that's kind of like do you believe that that means like leaders should kind of maybe go back to strategic books and business and then figure tech afterwards yeah in fact you know i actually i would you you know you go you you have to go several steps beyond that.

8:33You then have to figure out how do I, do I have the ability to actually execute on that strategy? Do I have the data? Do I have the talent? Do I have the investment, the stakeholder buy-in to be able to make sure that we can drive those solutions? So execution tends to be the main blocker for these types of things. But I also would suggest that people don't just assume that this is the smartest AI will be. You know, today is the dumbest AI will be and, uh, and it's just going to get smarter. So make sure that you're developing solutions that, that skate to where the puck is going on where the puck currently is.

9:12And now my prediction is, is that whilst, you know, a few years ago, large language models were a little bit like intoxicated graduates, you know, 50 % of what they came out with was nonsense. Um, and we're seeing that now graduating to like a master's level where they can do rudimentary reasoning. I think the fact is that over the coming years, they're going to go from master's level to PhD level to postdoc level where they can actually do science to eventually, you know, I predict that by the end of this decade, we'll have a professor in our pocket, something that we can actually ask questions that we've never asked and then use the scientific method to go and solve it.

9:47And I think that that's really where companies need to be building towards rather than where it is now. Yeah, so that's a great insight. So you've mentioned the need for strategy, execution, and arguably people, so culture, being able to adapt and embrace all this new change. So what guidance do you have for leading this space? Because we agree that technology is moving so fast. Today is the dumbest AI you'll use. Yet you have to work on operational efficiency, on increasing your customer delight. so you kind of have to do something today like right now so how do you balance out this dilemma between if i wait a little bit it's just going to get better so i could jump on that but if i wait i'm not really doing anything and my competition will get ahead so what's the trick here what's the recommendation so i think the trick is first you know educating yourself to make sure you're not being seduced by the hype or you know consultants that are just spouting rhetoric one-on-one.

10:48And so, so, and the second is, I know this is going to sound obvious. And the thing you've already said this, which is you sit down and you, you list all of the frictions that you have in your business. Frictions from your front office, you know, your, your supply chain to back office frictions to workforce frictions. And then you, you then list those frictions and prioritize them according to some simple business case. It doesn't have to be a complicated business case. And then they start knocking them off one by one, either, you know, buying solutions that solve that friction or building solutions um and then and then making sure that you're stitching this infrastructure together so that you have you know a an ai reusable ai architecture don't start with building a later a data lake and putting tableau on top and thinking that's going to drive value in your business make sure that you're addressing frictions and that's driving driving value and then and it's just it's an incremental process but we've done this with companies over the past you know almost two decades and and ultimately they end up driving value but then having an infrastructure that is scalable that's great so perhaps a bit more of an entrepreneurial sort of approach lower hanging fruits 2080 rule get started and then as you get some wins then you can like build the abstractions and the modularity around your initial work right that's kind of like the recommendation to get going um so don't wait but like start and approach it as a learning journey okay that's cool so if we talk about frictions and then making that whole list of friction what's the recent sort of ai project you've been involved where you know you've identified a team identified like here's a friction we think we can solve it you know with ai and what did you get out of it be great to get like an example to life i mean i mean if we look at marketing i think there were whilst there are many frictions across the kind of marketing supply chain there are there are seven really core problems one actually is creativity um how you can get ai to come up with ideas in divergent creative ways and uh and one of the things that we we've been doing in wpp is we've been training these ais um on the on the kind of expertise the genius of creatives and when we back test um those um the those AIs against griefs they've been able to come up with things that would have won can lions so so we can get AI to be creative um that said I believe they will totally enhance human creativity so it'll enable humans to elevate their creativity so that was one friction is how do you come up with more ideas faster better um content creation obviously is is is something that we do a lot but um creating brand specific production grade content is extremely hard if you can ask you know any online any model right now to go and create a brand perfect ad and it's not going to be able to even if you upload your brand guidelines your tone of voice your written copy imagery it's still not going to be perfect and and and so you know we've got a whole team of people that are trying to get to the point where ai is coming with brand specific be content.

14:01That's one friction. Another friction, which I think is really powerful that this applies to every company is you can build synthetic audiences of your, of who you're selling to. Um, and, um, you can then test content, whether they be sales materials, promotion materials, whatever, against those synthetic audiences to see how they think and feel about things. So, um, a generative AI has unlocked frictions that certainly across the marketing supply chain that now allow us to solve into a marketing, which wasn't possible before okay that's that's fantastic so there's a notion of you know the the ai can support a creativity process or even you know augmenting and some replace it to the same level you might might argue right it sounds like you had some sort of benchmark to to to identify whether the air solution is as good as maybe a team what about speed because that tends to be an underrated like value i imagine the space of marketing speed to creation is that like a factor like you react some events and you need to create stuff would that be like a benefit again historically that wouldn't have been possible because it would have been taken too long to come up with the ideas to be able to produce the content and push it out there by that by that point that moment has been missed and if i if i'd actually one of those frictions is the identification of macro and micro moments that are going on around the world right now where you could be building relevant campaigns to address you know that i'd argue that there are millions of moments today that are being missed where brands are failing to put their content in from the right people at the right time and and actually i actually think that that it's very exciting being in this industry because i think we're actually going to see way more marketing but much more relevant marketing and much more interesting um um marketing and when i say interesting is about you know if i if i asked you what your favorite ads were they weren't personalized ads they were probably you know an ad that had a shared cultural moment that connected people and that that sort of hit higher than or your values higher than just faster better cheaper but they made you feel nostalgic or connected or even make you feel like you're making the world better so i think what's really exciting over the coming years is that is that is it will be able to come up with these really crazy ideas that will connect you to brands in a much more quickly and an effective way so if i if i played back so you know make the list of sort of frictions and then context of marketing clearly creatives great place to look at speed plus the personalization so this is all the good stuff can you tell us a bit about the the challenges you know like again we can be sold on like there's so much value we can get it's going to be great but what does it take to make that happen in practice you know i'm thinking about you mentioned software system data like is it easy is it hard what does it take no i mean like any kind of transformation digital transformation you have to you have to take people on a journey and that's probably the biggest blocker is is going through that change management and marketing is going through, you know, complete change because historically you would sell time and materials.

17:19You would sell, um, your, your, your creativity hours. Now what we're having to do is sell and license enterprise AI software, which is a very different conversation with clients than just selling time and materials. So the business model itself is changing. Oh yeah. That's how, you know, the, the revenue model that in, of the entire industry is, is, is changing. and also going back to the point that you know people want to build a career for themselves so you know a lot of brands are then saying well we can build our own marketing tools now you know we've now we've got generative ai and again the reality is is that they will build tools they won't be sustainable they won't be differentiated and you know in three years time they'll have to go back hopefully to a wvp to to license our you know marketing software because it's smarter it's got all the data we've got teams of people are pushing the boundaries in terms of you know utilizing these models so i think you know the same the same barriers exist for you know any any emerging technology um it's just people placing them on bets and um and the change management required interesting at the end of the day you know pose on to people getting on the journey change management together so if we if we if we were to take that out and like where's the the defensibility in terms of you know you kind of mentioned there's players in the market how do you create defensibility you know in market like that whereby ai in a way is making things a lot more accessible you know like even there's a bunch of stories now that you no longer need a team of like 30 developers to kind of get an app out there you can like just go vibe code then maybe kind of tweak it a little bit so where do you see the defensibility is it the proprietary data set the expertise acquired over the years yeah i mean i think so defensibility will be in how how you train and pipeline these different technologies um so again if you wanted to create brand specific content doing that is very hard and um and we've got an entire team of people that try to figure out how to solve that problem that said you know it's going to become easier and easier to create more branded specific content but audience perception i think is a differentiator so if you've got data that helps you understand how people perceive things um i think um that that's more useful in your competitors than you have an advantage and wpp have just um launched open intelligence which is essentially a a human behavioral model that that helps us understand how people think and feel about things um so that that data i think is is is important you know purchasing data, knowing what's successful, what's not, you know, that's not available to everybody on the internet.

19:59So if you understand and predict performance, clicks, likes, sales, comments, because you have access to that data and then are able to understand why this is leading to clicks, likes and sales. And again, that is a differentiator as well. So I argue that there are really three things that differentiate a company in the world of AI. One is data, because it's data that makes AI smart. and marketing, big holding companies, marketing companies have a lot of data that is wall guard and that's proprietary. The second is an AI talent. I know most people have rebranded themselves as AI experts over the past three years, but the reality is if you want to build differentiated AI solutions, you need to have differentiated AI talent.

20:41And we're very lucky that Satalia was acquired by WPP. I think WPP is looking at that they acquired Stalia. But we have 300 deep experts that try to push the boundaries. And then finally, it's leadership. So I'll go back to what we talked about earlier. If leadership are being kind of not placed in the right bets, if they don't understand about the transformational power of these technologies, if they're getting seduced by the hype then then they make the wrong decisions and now is not the right time to making wrong decisions great love it so leadership on board educated whoever differentiated strategy people ai talent and proprietary data that's super so on the leadership point what's a misconception that business leaders have about ai that you wish you could like you know magically sort out well you know i do i do a lot of um talks trying to get business leaders to understand that there are different types of algorithms out there and ai is not just generative ai it's not a panacea to solve all problems across your organization um and so and so giving them a framework to to identify frictions and then apply the right technologies is i think it's something there's been there's been a very powerful tool that we've we've developed um so yeah it's just really going back to this idea of not getting seduced by the latest and greatest technologies and making sure that you're the that you're placing the right bets for the right technologies so if you have that conversation around like not being seduced by the hype how do you have the conversation around managing expectations you know it's like what can you get with air what can you not get and you know when you implement and i know safety is an important topic that is close to your heart so how do you manage all all of that you know it's like hey what's what's the right wait a minute you know ai is just software it's phenomenally powerful software because it can you know massively overachieve the goal that you've you've given it but just like software you need to make sure that you're testing it you're putting the right guardrails you're you're investing not just in the build of it but the maintenance of support innovating it so um so so i think that is is something that you need to remember if you want to build your own ai solutions there's 90 of stuff that often is missed that goes beyond just building it maybe a quick fire run of questions so i've got like a few short questions to ask what's the concern view you have in the industry i think we've got a we've talked about a couple but if there's like one that uh you know today you think hard about uh conferring view uh would be according to the definition of goal-directed adaptive systems that nobody's doing ai which is ridiculous comment to make but ultimately the true paradigm of ai systems that can safely adapt themselves in production and i think that's coming by the way cool and how do you stay up to date as a leader the world is moving so fast what's your way to like step today with everything happening well it's sort of my job to try and not get surprised by the future although that is becoming more and more difficult but i have a prediction that the ai will develop in this way that i've mentioned about kind of like a phd and postdoc and professor and i think about what are the implications of a world where these models are more and more capable i'm also interested in new emerging technologies from academia called neuromorphics which are models but from our brains work so the sort of hypothesis or ideas that is that the cost the energy cost the data required to train models is going to become exponentially less and so I don't I don't necessarily need to stay on top of all of the academic developments I just have to make sure that I'm sampling enough to make sure that I'm I'm I'm is validating my predictions great next is and I hope I know the answer to this one what's your favorite programming language well i've got a soft spot for java because i uh i learned that originally in uh in my undergraduate but i you know even matlab is a great uh tool to to hack about with um i think i expect most people nowadays will say python but um but i don't know i like i like java because that was where i learned about object orientated programming and when you think about learn about object oriented programming is again it's a way of thinking about systems and the interactions of systems so you need to be a systemic thinker um to be able to build um objectively uh objectory data programs that's that's crazy um because i agree with you uh i wrote a book called modern java in action it teaches object-oriented programming also functional programming i guess you probably like it because agents model as object and message passings model through method course so that's probably where it's come from.

25:51I was not expecting that then. I thought you'd say Python. What did you think I was going to say? Python. But yeah, Java is also my language, so that's great. And the nice thing about Java is there's abstraction layers and all this kind of stuff. So for me, it's a way of modeling actually the real world. Yeah, it's statically typed, so you're not getting code of code. What was your favorite subject at school? Oh, crikey. i was terrible at school i still am pretty bad at school um yeah i'm gonna answer the wrong question here but which is that it was philosophy so i actually learned philosophy outside of school that was a thing that i i was drawn to uh beyond beyond my studies um that said you know logic and reasoning and all that kind of stuff is fun but i guess that also links to philosophy as well yeah those are tightly coupled subject philosophy and logic amazing and final one what's your favorite music genre and you know i i was brought up by my grandparents in a small working class town in the corner of england and uh many many years ago and all they used to do is play like 60s music and i don't know also uh like musicals like andrew lloyd weber so i actually have a soft spot for for that for that genre uh unfortunately but um but i came into music late so i'm still catching up.

27:13Thank you, Daniel. It's been great chatting to you today. Nice one. Thanks.

27:21Fantastic conversation with Daniel. Super charismatic. Really enjoyed it. There's a couple of cool takeaways. The first one is clearly lots of hype in AI. Recommendation here is focus on frictions in your business. Literally make a list of all the frictions across all your business function supply chain marketing and so on and start tackling them you know one by one with a prioritization framework so forget the talk start working and uh get on with it and starting with frictions is the way to go the second really good insight was around the question of how do you create defensibility in this uh world of of ai there's three factors you need proprietary data that data is yours those are your insights you can utilize that to compete to you need ai talent people that can identify opportunities and build the systems that are going to deliver that competitive advantage and finally your leadership team needs to be on board and educated and recognize the opportunities and be willing to to fund them and if you go to three factors you can great defensibility in your organization and this world that is, you know, moving really fast.

28:39So thank you everybody and see you on the next episode. Thank you for tuning into this episode of Data and AI Mastery. If you found value in today's discussion, make sure to subscribe so you never miss an insight from the leaders driving the future of data and AI. And if you're a data and AI leader looking to upscale your workforce with the fundamental data and AI skills to transform your business, Cambridge Spark is here to guide you every step of the way. Be sure to reach out to us on LinkedIn or on our website, cambridgespark.com. Until then, be sure to keep pushing the boundaries of what's possible with data.

29:16And remember, mastery comes with continued learning and action. Until next time, stay ahead, stay inspired and stay masterful.